MadAppGang/magus

Full-platform browser automation. Live-page JS eval, keyboard and focus primitives, environment preflight, and a configurable agent LLM, across dual local and cloud browsers backed by Playwright's own Chromium so your real Chrome is never hijacked. Ten Magus-specific tools on top of the upstream Browser Use MCP set, plus 6 skills.

Bun and TypeScript toolkit. A `/bunjs:bun` index routes any task to the right one of eight task-shaped skills — setup, HTTP, errors, testing, security, production, performance, TUI — each shipping copyable code verified against Bun. Only Bun is tested.

MCP runtime for external AI models. Proxies OpenRouter, Ollama and LM Studio, runs blind team voting across models, manages async sessions, and streams channel notifications. Required by code-analysis, dev, multimodel, designer and seo.

Codebase investigation and semantic search. Maps architecture, traces callers and callees, and ranks symbols by PageRank through the mnemex AST engine, with a read-only detective agent for deep dives.

UI design validation. Compares a rendered screen against its reference by pixel diff, then reviews the result for spacing, hierarchy and design-system consistency.

Universal development assistant. Detects the project stack and routes work to specialist agents for architecture, implementation, debugging, review and research, enforcing design-system guardrails on every UI change.

BETA — Dingo language toolkit. Idiomatic Result and Option types, error propagation, enums, pattern matching, lambdas and safe navigation, with compiler-verified patterns. Interfaces may still change.

Go development toolkit. Ships the go-tui skill for building Charm-stack terminal UIs (Bubble Tea, Lip Gloss, Bubbles, ntcharts), with a colour-accurate screenshot workflow for reviewing a running TUI.

Getting Things Done workflow. Capture, clarify, organise, reflect and engage over a task store that hooks keep in sync with the session in real time.

Kanban board for task management. Five columns, cycle-safe task dependencies, priority indicators and WIP limits, over a task store independent of GTD.

Toolkit for madbench, MadAppGang's Go harness for benchmarking agentic coding tools. Ships the madbench-evals skill: author bench YAML and Eval files, design checks, scaffold red-state and generated testdata, run the negative and positive controls, tune expectations, and debug failing checks.

MCP runtime for semantic code search and AST analysis, ranking symbols by PageRank. Required by code-analysis and dev.

Multi-model collaboration and orchestration. Runs a task across several AI models in parallel and aggregates their blind APPROVE/REJECT votes, or delegates it to one model running a full Claude Code session.

Project setup jobs in one plugin. Investigates a repository and provisions it — plugins, tools, framework best-practice references, and a seeded knowledge base — then installs the adaptive statusline and indexes every skill reachable from the project into browsable markdown.

Deprecated redirect. The statusline now ships inside setup@magus as /setup:statusline-install, -uninstall and -customize. This plugin keeps the /statusline:* commands resolving for one release so no existing install breaks silently, and is removed after that.

Intent-level terminal control over tmux. Runs interactive commands, dev servers, test watchers, REPLs and TUI apps, and refuses to send keys into a pane that already has a process in the foreground.